Fixed regression with splitting out new subtree

A folder in a repository that wasn't initially imported as a  subtree could no longer be splitted into an entirely new subtree with no parent.

A fix and a new test to fix that regression is added here.

git-subtree.sh

@@ -253,6 +253,7 @@ find_existing_splits()
 				if [ -n "$main" -a -n "$sub" ]; then
 					debug "  Prior: $main -> $sub"
 					cache_set $main $sub
+					cache_set $sub $sub
 					try_remove_previous "$main"
 					try_remove_previous "$sub"
 				fi
@@ -569,7 +570,9 @@ cmd_split()
 		# ugly.  is there no better way to tell if this is a subtree
 		# vs. a mainline commit?  Does it matter?
 		if [ -z $tree ]; then
+			if [ -n "$newparents" ]; then
 				cache_set $rev $rev
+			fi
 			continue
 		fi
 

test.sh

@@ -294,6 +294,15 @@ git subtree split --prefix subdir --branch mainsub4
 # but it wasn't, because it's cache was not set to itself)
 check_equal "$(git log --pretty=format:%P -1 mainsub4)" "$(git rev-parse sub3)"
 
+mkdir subdir2
+create subdir2/main-sub5
+git commit -m "main-sub5"
+git subtree split --prefix subdir2 --branch mainsub5
+
+# also test that we still can split out an entirely new subtree
+# if the parent of the first commit in the tree isn't empty,
+# then the new subtree has accidently been attached to something
+check_equal "$(git log --pretty=format:%P -1 mainsub5)" ""
 
 
 # make sure no patch changes more than one file.  The original set of commits